Policy stories create a predictable spike in search demand and a predictable disappointment in the results. A minister is appointed, a rule is announced, a consultation opens, and within hours the same explainer exists on 200 marketing sites. Almost none of them are cited by the answer engines now handling the question.
The reason is structural rather than qualitative. On regulatory and policy queries, generative answer engines reach for primary and official material first, because that is where the authority for the claim actually sits. A well-written summary of a government position competes with the government position itself, and loses.
This article sets out why the explainer reflex fails on policy topics, what primary sources structurally cannot publish, and a method any team can run to measure the effect for its own subject area before committing a budget to it. The wider background is covered in our review of organic visibility in AI search.
What happens on a policy query
Answer engines decompose a policy question into sub-queries, resolve each from the most authoritative available source, then synthesise. Official documents, regulators and legislatures sit at the top of that hierarchy for questions about rules. A secondary explainer adds no authority the system needs, so it is frequently read and rarely credited.
Confirmed, on the decomposition itself. Google has described AI Mode as breaking a question into multiple sub-queries and searching across them rather than matching one string. That behaviour is what makes citation by answer engines a separate contest from ranking.
Measured, by third parties rather than by us. Ahrefs, analysing 540,000 query pairs, found AI Mode and AI Overviews cited the same URLs only 13.7% of the time. Moz found roughly 12% of AI Mode citations came from the classic top 10. A page can hold position 3 and be cited by neither of those answer engines.
The practical reading is narrow and useful. Being present in the ranked results is no longer evidence of being present in what answer engines return, and on policy subjects the gap widens because the primary source is unusually easy for a machine to identify and unusually hard to displace.
Why publishing the explainer first is now the weakest move
The reflex in most content teams is speed. A policy event lands, the brief goes out the same afternoon, and the piece explains what was announced. That instinct was correct when the contest was for a ranked position on a fast-moving term. It is close to worthless when the contest is for a citation.
The explainer is the exact shape of content that answer engines now satisfy without sending anyone anywhere. It restates a public document. It contains no claim that answer engines cannot verify faster at the source. It carries no information the primary text lacks, so there is no reason for a synthesis to reach past the original to find it.
Observed, and offered as practitioner judgement rather than as a measured finding. Teams that keep publishing policy explainers usually report the same pattern, which is respectable impression counts on the announcement term and very little else, because the piece answered a question that answer engines had already resolved above the results.
The correction is not to publish slower. It is to publish something the primary source cannot contain, which requires knowing precisely what it does not contain.
What official sources structurally cannot publish
A government document states a rule, a scope and a date. It does not state what compliance costs a 12-person firm, how the rule interacts with a different regulator's rule, which sector gets hit hardest, what a reasonable implementation sequence looks like, or what practitioners have found when they tried it. Those absences are permanent rather than accidental, and answer engines cannot invent what the record does not hold.
The United Kingdom's current AI position illustrates the gap cleanly. No standalone AI bill appeared in the 2026 King's Speech. The government instead announced a Regulating for Growth Bill placing regulatory sandboxes on a statutory footing, and has run policy through growth zones and existing regulators applying data protection, competition, equality and online safety law within their own remits.
Every one of those facts is published by the state, and the state will always publish them better. What no official page will tell a reader is which of those regulators is most likely to reach their sector first, what a sandbox application actually demands of a small firm, or whether a British business selling into Europe should follow the stricter regime by default and stop tracking two rulebooks.
Those are the questions with commercial weight. They are answerable only with judgement, sector knowledge and comparison, which is precisely the material answer engines must look beyond official pages to find. The political architecture behind this divergence, including why an AI brief now sits in Cabinet without a statute behind it, is examined in a companion report on the UK AI minister and how the job compares worldwide, published by Digital News.
A test you can run before committing a budget
The share of official sources cited by answer engines varies by subject, and no public dataset isolates it reliably by vertical. Rather than assume, measure it for the topic in front of you. The method below takes an afternoon and needs no paid tooling.
Assemble 20 questions a buyer would genuinely ask about the policy area, spread across the intents that matter, which are what the rule is, who it applies to, what it costs, when it starts and what to do about it. Put each question to the answer engines you care about, in a clean session with personalisation off, and record every cited domain rather than the answer text.
Then classify each cited domain into 4 buckets, which are official and regulatory, established news, trade and professional bodies, and commercial or vendor content. Count the share held by the first bucket, and record it per intent rather than only in aggregate.
The number that decides strategy is not the overall official share. It is the difference between the official share on the what-is-it questions and the official share on the what-do-I-do questions. Where that gap is wide, the second group is the only ground worth contesting, and the explainer budget belongs there instead. Where it is narrow, the topic may not reward editorial investment at all, which is a finding worth having before the commissioning rather than after.
Run it again after 90 days. Answer engines change faster than editorial plans, and a single reading tells you where you stand rather than where the ground is moving.
Where this reasoning stops
Uncertain, and stated as such. Whether the official-source preference on policy questions is stable, or an artefact of how these systems are currently tuned, has not been established publicly. A shift in either direction would change the arithmetic above, and anyone presenting a fixed percentage for answer engines as durable is overstating what is known.
The reasoning also does not transfer cleanly beyond regulated subjects. On questions with no primary authority, such as comparisons, methods and most commercial research, no document sits above the secondary layer, so answer engines fall back on the ordinary contest. Our earlier reading of how AI Overviews changed organic behaviour in UK search covers that wider pattern. Applying the policy playbook there would be a straightforward error.
Two further limits are worth naming. Semrush's January 2026 study found clarity and summarisation correlated with citation at +32.8% and promotional tone correlated negatively, which suggests format and register still matter to answer engines even where a topic is dominated by official material. The same study placed section structure at +22.9%, so how a page is organised remains part of the contest.
A subject can also be worth covering for reasons unrelated to being cited, including audience obligation and demonstrating competence to buyers already in a conversation.
Disclosure, since it is relevant to the argument. Teksyte sells search and content services, so a piece recommending where editorial budget should sit is not disinterested and should be weighed accordingly.
What this changes on Monday is the brief rather than the calendar. On any policy-adjacent subject, stop commissioning the explanation and start commissioning the application, which means cost, sequence, sector consequence and the comparison nobody official is permitted to make. Run the citation-share method first so the decision rests on the topic in front of you rather than on a general claim about answer engines.
What remains genuinely open is durability. The official-source preference could soften as these systems mature, and no public dataset yet tracks it over time by vertical. Review date for this article is 3 months, which is the standard interval here for anything touching AI search behaviour.